Mini Demo 3: Public Health & Epidemiology
Merrimack College DSE6630: Healthcare & Life Sciences Analytics
Adeline Casali, Samantha Harper, Scott Eugley
3 June 2024
Epidemiology of pneumonia-related hospital readmissions
As you work through your Project 2, keep in mind that we are actually working with spatial data even though we have not focused on that much until now. This relates to everything we will be doing in Module 3 on Public Health & Epidemiology, though! As data scientists of public health and epidemiology, one of the most powerful tools in our toolkit are maps.
For our hospital readmissions data, we had a lot of spatial data that we could have leveraged but didn’t really in our other analysis. This demo, we are going to take advantage of the fact that there is a LOT of spatial data here that we can use: addresss, county, state, zip code… wow! We will focus first on the state level data together, and then I am going to have you try your own analysis at the county-level.
1 State-level Analysis of Pneuomonia-related Hospital Readmissions
1.1 Make a base map.
Let’s practice bring up a very simple little map of the US from
maps and ggplot2. The maps
package provides latitude and longitude data for various in the package.
We will do something much more complicated for our Project 3, but for
this little exploration let’s just keep it simple.
Notice that first I am having to extract state-level geographic
information from the map_data() function. Similar functions
exist for other geographic levels of data, including world (country),
county, or other regions of the world.
Great, our first map of the continental US! But it’s pretty uninteresting just as it is. Let’s use this map to overlay our pneumonia-related hospital readmissions data.
1.3 Summarize the data by mean
PredictedReadmissionsRate.
## Grab just the readmissions and the region, drop the NAs, then group by region and calculate the mean readmissions
avgReadmissions <- pneumoniaFull %>%
select(PredictedReadmissionRate, region) %>%
drop_na() %>%
group_by(region) %>%
summarize(avgReadmit = mean(PredictedReadmissionRate, na.rm = TRUE))| region | avgReadmit |
|---|---|
| district of columbia | 18.39 |
| nevada | 17.94 |
| west virginia | 17.69 |
| california | 17.65 |
| new jersey | 17.63 |
1.4 Merge mean PredictedReadmissionsRate with the
states data for mapping.
mergedStates <- merge(avgReadmissions, states, by = "region")1.5 Map!
Do you notice any trends that would be worth testing? For example, do you see any pattern with very populated states?
2 Assessing other state-related spatial correlations
Perhaps you noticed, as I did, that there seemed to be higher rates of pneumonia-related readmissions in states with higher population sizes, like California, Florida, and New York (although Texas is not among them). Since admissions have been measured as a percent of pneumonia cases admitted to the hospital, they should not in theory be sensitive to population size beyond what is happening in the surrounding environment. In other words, we don’t need to worry about scaling or adjusting these values because we are looking at a percentage. However, there could be underlying associations or correlations that are worth exploring.
- Hypothesis 1: States with larger populations have higher average pneumonia-related readmissions because populations are more dense, enabling both pneumonia spread and are seen in states with larger populations.
But we had also posed back in Demo 2 that perhaps it has even more to do with the population of elderly individuals, as these individuals were most suspceptible to death by pneumonia for various age-related reasons. Perhaps higher readmissions has more to do with who is being infected with pneumonia in addition to the qualities of the hospitals.
- Hypothesis 2: States with a higher proportion of their population over the age of 65 have higher rates of readmission because of age-related factors that increase the likelihood of severe illness and even death.
But these are not the only population-related hypotheses we could pose. Perhaps, instead, it has to do with poverty - perhaps states with higher poverty rates also tend to see lower rates of medical coverage, thereby leading hospitals to release patients too early to spare costs. Thus, we might also ask if states with higher poverty rates are more likely to see higher pneumonia readmissions?
- Hypothesis 3: Higher rates of pneumonia-related readmissions may be associated with higher rates of poverty because of poorer medical coverage, including non- and under-insuredness, which in turn could lead some hospitals to turn out patients sooner than they should.
2.1 Pre-processing the population-level data to test hypotheses
Let’s bring in some data from the 2023 US Census (these are
projections based on the 2020 Census), as well from the Administration
for Community Living (ACL) which had conveniently already compiled data
from the US census bureau for our hypotheses 2 and 3. The 2023 US Census
Bureau projections were downloaded from here,
and the ACL data were downloaded from here.
For our convenience, I already put them into a CSV we can
import. The geographic measurements also come from the US Census Bureau,
which I obtained here.
## Read in, calculate the percent of the total adult population over 65,
## Make the states lowercase
## Calculate population density as hundreds of people per square mile
pop <- read_csv("2023_US_Census.csv", show_col_types = FALSE) %>%
mutate(over65PercentPop = over65Pop / adultPop,
region = tolower(region),
popDensity = (totalPop/100)/(landSqMi)) Notice that I calculated the % of the population that is over the age of 65 relative to the total adult population, but how should we best address Hypothesis 1? We could just use the total population size, but since our question has to do with population spread, we would be better served to calculate the population density. Higher densities would facilitate faster spread; lower densities, lower spread, at least by conventional wisdom. Thus, we do that here as well!
| region | totalPop | adultPop | over65Pop | percBelowPoverty | landSqMi | over65PercentPop | popDensity |
|---|---|---|---|---|---|---|---|
| connecticut | 3617176 | 2894190 | 663318 | 0.07 | 4842 | 0.23 | 7.47 |
| district of columbia | 678972 | 552380 | 87260 | 0.19 | 61 | 0.16 | 111.31 |
| massachusetts | 7001399 | 5659598 | 1260538 | 0.11 | 7800 | 0.22 | 8.98 |
| new jersey | 9290841 | 7280551 | 1611767 | 0.07 | 7354 | 0.22 | 12.63 |
| puerto rico | 3205691 | 2707012 | 756849 | 0.40 | 3424 | 0.28 | 9.36 |
| rhode island | 1095962 | 892124 | 206410 | 0.07 | 1034 | 0.23 | 10.60 |
2.2 Hypothesis 1: Readmissions by Population Density
2.3 Hypothesis 2: Readmissions by Percent of Population 65+
2.4 Hypothesis 3: Readmissions by Percent of Population Living Below the Poverty Line
Question 1 (SE) [2 points]:
Which hypothesis is supported, if any? Make sure to briefly discuss, in a sentence or two, each hypothesis and the evidence you see here. Does it warrant any kind of follow-up?
Hypothesis 1: This hypothesis does seem to be supported, based on the correlation scatterplot. There is a positive correlation between pop. density and pneumonia-related hospital readmission. States with higher pop. density tend to have higher readmission rates. Some follow-up is warranted, I would like to look at the density/readmission relationship at a finer geographic level, like county or by zip code. This could help identify specific areas of high density that could be driving this correlation.
Hypothesis 2: There does not seem to be a clear correlation between % of adults over 65 and readmission rates across states. There still may be some options for further explanation before giving up on this hypothesis. Perhaps investigating potential non-linear relationships, or look at different age groups, like 70+ or 75+ to see if there’s a correlation in the oldest age groups.
Hypothesis 3: There is a possible positive correlation here, but it is not strongly supported.This hypothesis could also benefit from looking at it from a finer geographic level, as poverty can vary greatly within states.
3. County Level Analysis of Pneumonia-related Readmissions
Now, it’s your turn! We’re going to practice our new mapping skills by having you practice working with the county data.
3.1 Make a base map.
Use the maps package and the map_data()
function to extract the county level data and plot a base map:
Question 2 (SE) [2 points]:
I get you started a little bit below, but now it’s your turn to try to make a base map of the US counties.
3.2 Extracting county-level data for use
zipcodeR is an incredibly handy package when working
with US data! The search_state() function will look up
information we need based on the states we have at the city- and
county-levels, which we can use with the
## Apply the search_state function to get all sorts of wonderful zipcode level details
stateInfo <- search_state(pneumoniaFull$State)3.3 Summarize the data by mean
PredictedReadmissionsRate.
Question 3 (SE) [3 points]:
Summarize the county-level data to calculate the average pneumonia
related hospital readmissions so that you can merge this with the
stateInfo table above as well as the counties
dataframe that we extracted from map_data().
Hint 1: My code above should be an excellent starting point if you need it.
Hint 2: You may find it easiest at this point to go
ahead and rename your column CountyParish to match the
counties data frame! Also make sure to make it lower
case!
Hint 3: You are still going to need the state
information from the readmissions dataset! You will want to group by
both region AND subregion!
Hint 4: You will want to remove the NA
data before merging, but I would suggest doing it right before you group
and summarize.
# Rename 'CountyParish' column and make lowercase
pneumoniaFull <- pneumoniaFull %>%
rename(subregion = CountyParish) %>%
mutate(subregion = tolower(subregion))
# Summarize data by mean PredictedReadmissionsRate at the county level
county_readmissions <- pneumoniaFull %>%
select(PredictedReadmissionRate, region, subregion) %>%
drop_na() %>% # Drop NAs
group_by(region, subregion) %>% # Group by region and subregion. Then summarize.
summarize(avg_readmit = mean(PredictedReadmissionRate, na.rm = TRUE))## `summarise()` has grouped output by 'region'. You can override using the
## `.groups` argument.3.4 Merge mean PredictedReadmissionsRate with the
counties data for mapping.
Question 4 (AC) [1 points]:
Merge the counties data with the summarized readmissions
dataset you just made.
Note: I got you started here because there’s a bit
of trickery. You must arrange by the order
(which tells the polygons what order to plot in). Merging will jumble
them up and plot a big mess if you don’t order them!
# Merge the counties data with the summarized readmissions dataset and arrange by order
counties_with_readmissions <- merge(counties, county_readmissions, by = c("region", "subregion"), all.x = TRUE) %>% arrange(order)Question 5 (AC) [3 points]:
Merge the counties data with the summarized readmissions
dataset you just made.
plot <- ggplot(counties_with_readmissions, aes(x = long, y = lat, group = group, fill = avg_readmit)) +
geom_polygon(color = "black", alpha = 1) +
scale_fill_continuous(name = "Mean Readmissions", low = "white", high = "#01387a", na.value = "white") +
theme(
panel.grid = element_blank(),
panel.background = element_rect(fill = "white"),
plot.background = element_rect(fill = "white"),
axis.text = element_blank(),
axis.ticks = element_blank(),
axis.title = element_blank(),
legend.position = "bottom"
) +
ggtitle("Average Pneumonia-Related Hospital Readmissions by County")
# Print the map
plot
Question 6 (AC) [1 points]:
What do you notice immediately about the county-level data? Why do you think this is an issue? Are there some counties that legitimately lack hospitals entirely?
The first thing that I notice about the county-level data is how cluttered and conjested it looks. It’s incredibly hard to notice trends when the counties are clustered so closely together, particularly in the middle of the country. Additionally, some counties may not have an acute-care hospital, or a hospital altogether in extremely rural areas of the country, further skewing the results of the visualization.
Question 7 [5 bonus points]:
Use the stateInfo dataframe to explore a
hypothesis of your own about county-level hospital
admissions. You could look at population density again, median income,
median home value, or even the number of housing units!
Hint: You would need to merge the data again with
the counties dataset so that you could make a map.
# Your code here